Extracting Features from Bank and Mortgage Company Commercials

Abstract

Recent advances in machine learning have made it possible to extract rich data features from video. These techniques allow for a new understanding of the role of advertising in marketing products. I demonstrate how to extract text, image, and audio features from a novel dataset of over 20,000 bank and mortgage company commercials covering the years 2004 to 2015. I demonstrate that the extracted features display meaningful trends. I then apply these features to show how commercials impact mortgage and home equity loan origination and refinancing.